Hacker News new | ask | show | jobs
by midway 2690 days ago
What I don't get: Either you go full SPA or SSR. And If you go for latter then you can choose any stack, e.g. Rails, express with pug, Django, PHP, etc. Why do people take React's constraints to the backend? No bashing, really wondering what's so much better then. I know why a React SPA can be better than a SSR. But what is better about a React SSR vs the typical SSR?
10 comments

Simple, SEO. We want full SPA, but also want SEO to work without shaky unreliable prerendering. Hence server-side-rendering for SPAs.

Some will tell you that it's for first time loading performance, but in this day and age, I doubt that even registers for business impact.

Still don't get it. Next.js gives you a vendor lock-in and its features are easily avail w/o Next (eg. automatic code-splitting, this is standard stuff in CRA, why the fuss?). If your business is about SEO then React is the wrong choice anyway. In my understanding SEO is about hundred-thousands or even millions landing pages, keyword management, content spinning, ultra fast loads, AMP, etc. This is really not the field of React or SPAs and you can tweak them as much as you want. React is great to build eg stateful SaaS, interactive stuff, dashboard, nice shop check-outs.

That they rewrite Next doesn't sound more convincing. And why do I need to sign-in on the learn page on Nextjs?

Sorry, not convinced. If you want some brute-force SEO, I'll write you anything in express/pug/style in one afternoon (with millions of landing pages, sitemap generation and smart routing, all comes out of the box with express), without all the baggage React brings along. Nothing against React, it's still one of the best SPA libs there but SSR, no way.

Maybe, I am still missing something but Next.js feels just like content marketing from its company Zeit trying to promote their commercial stuff.

Edit: guys, pls don't downvote if you disagree, downvoting is if comments don't add anything substantial to the discussion. I understand that Zeit is happy about the free promotion for Next on HN and doesn't like such comments but pls stay professional if people question your product without downvoting.

I’m very involved in this space. The company I work for wrote a library similar to Next.js before it was announced and before create-react-app. There are so many things that these solutions solve to only focus on SSR is missing the point. SSR is one benefit of many and others have already explained why they care about that. There are many use cases where both SEO and complex interactions make sense on the same page. Next.js is also going to help you with overall architecture decisions like routing, page level code splitting, where you fetch data just off the top of my head. Sure you can build these yourself, that’s what I did before Next.js was a thing but now that I have a deep understanding of all of the things we wrote. I would use an existing community backed solution before building from scratch in a new app. We are even very interested in migrating from our home rolled solution. Currently we are vendor locked into our own creation which means we have to continue to maintain it. It is way more expensive than any lock in that would have come with Next.js.
+1, even though I don't use the next.js tools, sometimes having certain things configured in the box is useful. I tend to stick with jest for testing simply because code coverage tooling can be a pain to get working right with babel, etc and it's just easier to use jest.

As to why SSR React, it's mostly going to come down to SEO, I don't know how much you'll get negatively impacted for being SSR vs SPA these days with proper routing myself. Google has been working on supporting client rendered scraping for at least 6-7 years now. And when a user hits more than one route, they've usually gained back anything they've lost for the extra load time.

> Next.js is also going to help you with [...] routing

I respect your experience and knowledge but people in this thread say exactly the opposite: Next.js' routing is actually not helping you.

Fair enough. It helps you but it could be better. Depends if you need dynamic routing or not. Personally I don’t find their solution for dynamic routing terrible but it could be better.
I don't buy the vendor lock-in argument here. The vast majority of your code will have nothing Next-related in it, this is a sharp contrast to a lot of the other options out there. Your page entry points will have getInitialProps, but that's actually a pretty decent pattern I'd probably choose to keep even if I moved away from Next.

The reason we actually went with Next is because it's relatively easy to drop if our needs diverge.

Of course there is a lock-in. Every lib you use locks you in. You don't know how many libs don't play well together. The more code and libs you add to your app the more complexity.

It's not hard to write code, it's hard to keep code complexity low and maintainable and IDK if Next.js helps here.

Yes, all libs give you vendor lock-in, which means unless Next is a particularly egregious example, there's not much point calling it out as a particular wart. My argument is that it's closer to the opposite, because as far as web frameworks go, Next does this better than most -- very little of the code you write is different just because you're using Next, your code is just normal React code. This is a huge contrast to something like Gatsby, where the lock-in is much stronger.
Sorry but I disagree. Just take any of your a bit bigger React projects and transform them to Next or the way around. In. One. Hour. And production-ready please. Good luck, man (this is vendor lock-in).
SEO isn't just about landing pages or corporate websites. Every public facing route is part of SEO.

Some of the most SEO heavy websites are social networks, ecommerce sites, marketplaces and other interactive websites.

Even for corporate websites though, why should we degrade the user experience for SEO if we can easily have both?

And yes whether you agree or not (and by your comment I think you don't) the vast majority of users and UI/UX designers consider SPA to be a much better user experience than loading pages.

If you are making websites for a living, I suggest you come aboard the SPA train quickly or risk losing north of 90% of job opportunities.

> Next.js gives you a vendor lock-in and its features are easily avail w/o Next

Think Next.js not like a framework but an SSR version of create-react-app. It's incredibly easy to port into your custom thing.

If you're already on create-react-app or any other framework or stack and don't want to port to a new framework, but want to add easy "just works" SSR, you can use react-frontload.

https://github.com/davnicwil/react-frontload

It's a small library that lets you load data into Components - using the same code on both client and server render. You can use it to add server rendering to any existing React app. It's designed to just plug in and work. I'm the author :-)

Pls check my other comment, anything you add to your code adds complexity and a natural lock-in. Maintainable code is not about adding every lib which crosses your way.
I see Next.js more as a build system than a framework / library. Of course it adds complexity, but in a comment above you say that "CRA does that too". You're not introducing any more or less complexity than CRA.
And what I'm saying is that Next.js does add the same complexity that you would add yourself but with great documentation. Also I have to say that's a very generic argument for any library.
Then why don't you just use the automatic code splitting of CRA?? And it's not generic, if you ever worked on a years old code base you are happy if complexity is low. Do you know if all the page transition libs will play well with Next? They do not play well even with react-router, the standard React routing lib. Code can be more complex than you might think.
CRA is a SPA. Next.js is not. When you hit a page in Next.js, it can prefetch and hydrate the data before it sends it down to the client. CRA cannot do this. First the client has to load the app, then the app has to fetch data from the server.
Next.js gives you prefetching as well as code splitting as well as server rendering.

You get to use all the modern tools that React gives you, an abundance of open source React components, and get an SEO and load time optimized website with minimal configuration. Building a server rendered, dynamic web app is non trivial. Next.js makes it a lot easier.

React coupled with Node.js has been used for SSR since 2015 in sites like PayPal so I'm not sure why you doubt it's veracity. React is a very strong UI library that is more than suitable for multiple rendering targets. It has never been limited to SPA apps. Facebook itself is not a pure SPA app.

> prefetching

Prefetching is a questionable and a lot discussed feature, do you want to prefetch all the 50 links on my page? Even if you think this a great feature, I am sure the user using your page doesn't like that you abuse his/her bandwidth. Especially those on mobile.

> code splitting

This is not a feature. Automatic code splitting is turn-key-ready in CRA. I wrote this now for the tenth time...

> server rendering

I explained this, if you really need SSR, a plain express/pug blows Next out of the water. Speed, flexibility, routing, everything. There is no single reason to use anything like Next.js. And writing a plain express/pug app is trivial and much easier than React, this is 10 years old tech.

I don't want to sound negative but again why is React as SSR 10x better than typical SSRs? It's not, React is great for real SPAs and Next feels to me like, let's jump on the React hype train and generate organic leads for Zeit.

You're just being nitty and not adding any value to the conversation.

Code-splitting in Next.js is fundamentally different than in CRA: https://nextjs.org/#automatic-code-splitting https://facebook.github.io/create-react-app/docs/code-splitt...

It's automatic, handled server side...and it gets hydrated on the server side if necessary with asynchronous data. The benefits of this on browser performance and developer performance should be self-evident.

Doing a truly isomorphic React app on your own with pure Node.js is non-trivial. As for using pug templates...just...no. You might want to read about why JSX is a thing to begin with.

It's not a hype thing. It's a tool that solves a problem. It's not needed for all applications. For many apps, a SPA is fine. That an engineer has to make evaluations on the trade-offs of various decisions goes without saying...But Next.js is a solid choice for a SSR framework.

Next.js is some over-engineered thing without a proper use case.

If Next.js was good in SEO, nextjs.org would rank well on the search query 'react ssr' but you know what, it couldn't even rank within the top 10. There's nothing, no nextjs dot org, just a Medium article below the fold. Just want to see your face after you convinced your management and your entire team to move to Next and facing how your organic search volume is nosediving...

This whole thing is pure content marketing paired with forum spamming to get people converted to Zeit's products. I am bit surprised that not more 'real users' defend Next.js here, seems like just some maintainers/contributors could show up worshipping Next. This is all too obvious and maybe young devs fall into your trap.

Btw, I know what JSX is capable of and I prefer it over many things but it's not about JSX or pug or whatever. Next.js is abusing a powerful lib/framework for SSR and templating. SSR is much more, e.g. proper routing, read the thread, Next doesn't get routing (the #1 req of SSR) right and wants to do SSR? Sure.

> Prefetching is a questionable and a lot discussed feature, do you want to prefetch all the 50 links on my page?

This isn't how Next prefetches with dynamic routing. It fetches the JS required to construct the DOM of those pages, but doesn't execute it. If on one page you have three links to /gallery/a, /gallery/b, and /gallery/c, then gallery.js will be prefetched once. You can confirm this in devtools. Also, prefetching is opt-in.

Forget about nextjs and SPAs for a moment, think of react as a better alternative to pug, instead of express/pug you can use express/react, better because is component based and you write all your logic in JS instead of some new limited mini language.
I will sacrifice more my HN points since I will add nothing but to say that regardless of rationality you put, if being negative to React/javascript, it's guarantee to get downvote to hell.
I don't think it's any better or worse, it's just an additional way of writing SSR sites where React is a templating language.

React is just another option for a stack - similar to express with pug, you can have express with React, without having to learn a new templating language if you already know React. And with things like TypeScript, you can have type checking into the mix too

I selected React & Next.JS for our templating framework to use with an API driven website. There are hardly any SPA features incorporated into the website, though we set that as a possible down the road feature.

For me, the biggest reason to use React & Next.JS over any other template language was how enjoyable the JavaScript ecosystem is to work with once setup. React just happens to be the modern day JQuery & is leading the pack. I also prefer its XML/JSX type syntax.

The ability for us to write code that would work on the SSR portion & front end was also a big plus.

Also Storybook isn't a React only thing, but it is a JS thing. Allowing our designers to create components with mock API calls & having Webpack update the component (without reloading) on save is an amazing time saver & productivity gain. I know of no other language that has a component template system that is close to JavaScript, Storybook & Webpack. As WASM matures, I hope to see other languages match it.

Because they already know React, and they don't know Rails, Pug, Django, Laravel, et al. And using the skills they already have is easier than learning new ones. When all you have is a hammer, everything looks like a nail.

It's the same reason why Electron is so popular for desktop apps. On just about every measure, there's a better alternative available for that sort of thing -- native apps if you want speed and great integration with the host environment, Qt if you're willing to sacrifice on those fronts to get a single cross-platform solution. But Electron has one thing those alternatives don't, namely that if you're a Web developer you don't have to learn any new languages, toolkits, etc. to be productive with it. And it turns out that, for an awful lot of developers, "I don't have to learn anything new" beats just about every other consideration.

React's constraints aren't particularly bad... the context bases for controls in a single file, better with React-JSS imho make it much easier to write against. SSR with React is sometimes a PITA, but that's rarely required imho.

I'm okay with SSR, but often you get to a point where you want to mix in dynamic functionality on the client that just becomes more difficult with it. I think that Vue, React and others can be integrated into SSR, but doing SSR of React makes for a nice application development as a whole imho.

In the end, it's up to the developer. I will say with JS client and direct server there's less disconnect for most development, and the lack/reduction of context switching can be a huge productivity boost. I'm usually a proponent of a JS client talking to/through a JS API, that can talk to other things. SSR with that server platform just makes it a bit more coupled, which isn't in itself a bad thing.

YMMV.

With frameworks like Next.js, it's no longer either/or, the explicit goal is to have the best of both worlds. The first page is delivered with SSR (for speed), then it boots React in the browser and works as an SPA.
I still think that you should endeavor to server-render anything you can get away with. Because of that, we have invested heavily in React SSR.

The biggest value-add of using React on the server is being able to share our design library and business logic with our front- and back-end. We can develop the structure, style and behavior of our website as React components and deploy them server side or client-side. In general, our team has found that the component model is far more composable than templates like pug, PHP, etc.

> And If you go for latter then you can choose any stack, e.g. Rails

You suggested Rails, but a big reason people like Next is exactly because it's an opinionated, batteries-included framework like Rails.

The "vendor lock-in" situation, which seems to be one of your main arguments, is exactly the same time. If you don't like vendor lock-in, I don't know why you'd propose Rails.

Next.js is a SPA with that loads a little faster and is more SEO-friendly.
> Next.js [...] loads a little faster

IDK, my dockerized express site loads much faster. At 20ms without network time (and this w/o caching). My real React SPA give you instant (0ms) page loads.

Nextjs.org initial load is at 580ms and the doc came at whopping 1.38sec (oh yeah) and page loads are not instantly... this thing doesn't make sense (for my use cases). If the only thing I know is React and JSX, then maybe but routing with express is much easier.

=> you want make serious money with SEO => get some real SSR environment

=> you want a great SPA, buttersmooth, interactive, maintainable code => React or other popular SPA libs

How can that be possible? The DNS lookup takes longer than that and you still haven't even gotten to your site. Once there the first byte time from the initial HTTP round strip still take considerably more. This is all network traffic that does matter what the application is doing.
> How can that be possible? The DNS lookup takes longer than that and you still haven't even gotten to your site.

I wrote without networking time which includes DNS lookup

How long is the first page load of your react SPA?
This is the wrong question. I use React for SPA use cases and then initial loading time is secondary but I tell you something, my stock React SPA with tons of js and assets has following loading times:

  DOMContentLoad: 1.08sec, Load: 2.26sec
and let's look at nextjs.org:

  DOMContentLoad: 1.09sec, Load: 2.31sec
I mean this comparison doesn't make sense because you don't know my SPA and I could write anything here but test it yourself. I just don't see a clear use case for Next.js. Just make an example, you sell some SaaS. Make the web product as React SPA but the marketing landing pages as normal express/pug site. Both the web product and the landing pages have different use cases: one needs to create leads and the other one has to increase customer satisfaction. These are different goals, so maybe you should also use different tools.
My own react app, not using next with a rather heavy payload (async splitting in webpack broke in the past 3 months and haven't been able to get it working again) is under a load of 1.24s Finished time, locally (static content server, with 2 db queries via api).

Unless you MUST have the absolute best SEO, I'm not sure that it really matters having SSR for most people/apps. I don't know how much traffic is really driven from search engines depending on the site, and I really don't know how much you get penalized for having an SPA anymore. I mean if your initial load is under 2-3s and in-site navigation is fast, you'll probably be fine.

The past few sites/apps I've consulted on got the majority of their traffic from outside SEO.

DOMContentLoaded isn't as relevant with SSR, because the markup and styles come rendered on the first response, so you never see an initial blank white page.

More relevant metrics are TTFB (time to first byte) and TTI (time to interactive).

I wouldn't want to implement a marketing site in express/pug. I want the flexibility of components and all the other benefits that React offers.

> How can that be possible? The DNS lookup takes longer than that and you still haven't even gotten to your site.

Me too. But if you ever did marketing landing pages for products you might know that one landing page is not enough. You need thousands which match to Ad Word campaigns or which are optimized to different SEO fields, domains and whatever. While you still could manage this with Next and nice components, Next's subpar routing might be an obstacle.

SPA? SSR?
SPA = Single Page App SSR = Server Side Rendered

SPAs run in the browser, therefore they must be written in JavaScript (or something that compiles to it.)

SPAs have the drawback of having to load a bunch of JavaScript before rendering anything, so the initial page load tends to be slow. Next.js gets around this by performing SSR for the initial page view, and then you're already looking at that page while all of the JS loads to render subsequent pages.

The term 'SPA' is a little outdated - since HTML5 history (which is nearly a decade old now) most 'SPAs' are actually multiple pages.

'web apps' is better.

They stand for Single Page Application (client side) and Server Side Rendering (which makes a JS app render more like a normal server side page for the client) respectively.
single-page-app, server-side-rendering